Discussiones Mathematicae

Graph Theory 41 (2021) 697–711

doi:10.7151/dmgt.2347

GRAPH OPERATIONS AND NEIGHBORHOOD

POLYNOMIALS

Maryam Alipour and Peter Tittmann

University of Applied Sciences Mittweida
e-mail: malipour@hs-mittweida.de
peter@hs-mittweida.de

Abstract
The neighborhood polynomial of graph G is the generating function for
the number of vertex subsets of G of which the vertices have a common
neighbor in G. In this paper, we investigate the behavior of this polynomial
under several graph operations. Specifically, we provide an explicit formula
for the neighborhood polynomial of the graph obtained from a given graph
G by vertex attachment. We use this result to propose a recursive algorithm
for the calculation of the neighborhood polynomial. Finally, we prove that
the neighborhood polynomial can be found in polynomial-time in the class
of k-degenerate graphs.
Keywords: neighborhood complex, neighborhood polynomial, domination
polynomial, graph operations, graph degeneracy.
2010 Mathematics Subject Classification: 05C31, 05C76.

1. Introduction

All graphs considered in this paper are simple, finite, and undirected. Let G =
(V, E) be a graph where V is its vertex set and E its edge set, suppose v ∈ V
is a vertex of G. The open neighborhood of v, denoted by N (v), is the set of all
vertices that are adjacent to v,

N (v) = {u | {u, v} ∈ E},

and the open neighborhood of U , for a vertex subset U ⊆ V is

[
N (U ) = N (u) \ U.
u∈U
698 M. Alipour nad P. Tittmann

The neighborhood complex of a graph G, denoted by N (G), has been introduced

in [15]. It is the family of all subsets of open neighborhoods of vertices of the
graph G,
N (G) = {A | A ⊆ V, ∃ v ∈ V : A ⊆ N (v)}.
The interested reader is referred to [15] and [5] for more properties of the neigh-
borhood complex N (G).
The neighborhood polynomial of a graph G, denoted by N (G, x), is the ordinary
generating function for the neighborhood complex of G. It has been introduced
in [5] and is defined as follows
X
(1) N (G, x) = x|U | .
U ∈N (G)

Suppose |V | = n, and let nk (G) = |{A | A ∈ N (G), |A| = k}|. Then we can
rephrase the Equation (1) as follows
n−1
X
N (G, x) = nk (G)xk .
k=0

The neighborhood polynomial of a graph is of special interest as it has a close

relation to the domination polynomial of a graph. A dominating set of a graph
G = (V, E) is a vertex set W ⊆ V such that the closed neighborhood of W is
equal to V , where the closed neighborhood is defined by

N [W ] = N (W ) ∪ W.

We denote by D(G) the family of all dominating sets of a graph G. The domi-
nation polynomial of a graph, introduced in [3], is
X
D(G, x) = x|W | .
W ∈D(G)

For further properties of the domination polynomial, see [1, 6, 7, 12, 13]. It
has been observed in [4] that a vertex set W of G belongs to the neighborhood
complex of G if and only if W is non-dominating in the complement (Ḡ) of G,
which implies

(2) D(G, x) + N (Ḡ, x) = (1 + x)|V | .

A proof of this relation is given in [10].

In Section 2 of this paper, we investigate the effect of several graph operations
on the neighborhood polynomial of a graph. The vertex (or edge) addition plays
an important role in this context.
Graph Operations and Neighborhood Polynomials 699

In Section 3, we present a recursion formula for the neighborhood polynomial

of a graph based on deleting vertices. We apply this recursion to several graph
classes and prove that the neighborhood polynomial of planar graphs can be
computed efficiently.
In Section 4, we investigate the complexity of calculation of the neighborhood
polynomial of a graph and finally in Section 5, a conclusion as well as some
interesting open problems are discussed.

2. Graph Operations

Having defined a graph polynomial, one of the first natural problems is its cal-
culation. Often, local graph operations, such as edge or vertex deletions, prove
useful. In addition, global operations, like complementation or forming the line
graph, might be beneficial. Finally, graph products, for instance the disjoint
union, the join, or the Cartesian product, can be employed to simplify graph
polynomial calculations.
The disjoint union of two graphs G1 = (V1 , E1 ) and G2 = (V2 , E2 ) with
disjoint vertex sets V1 and V2 , denoted by G1 ∪ G2 is a graph with the vertex set
V1 ∪ V2 , and the edge set E1 ∪ E2 .

Theorem 1 [5]. Let G1 and G2 be simple undirected graphs. The disjoint union
G1 ∪ G2 satisfies

N (G1 ∪ G2 , x) = N (G1 , x) + N (G2 , x) − 1.

The join of two graphs G1 = (V1 , E1 ) and G2 = (V2 , E2 ) with disjoint vertex
sets V1 and V2 , denoted by G1 + G2 , is the disjoint union of G1 and G2 , together
with all those edges that join vertices in V1 to vertices in V2 .

Theorem 2 [5]. Let G1 = (V1 , E1 ) and G2 = (V2 , E2 ) be simple undirected

graphs. The neighborhood polynomial of the join G1 + G2 of these two graphs
satisfies

N (G1 + G2 , x) = (1 + x)|V1 | N (G2 , x) + (1 + x)|V2 | N (G1 , x)

− N (G1 , x)N (G2 , x).

The Cartesian product of graphs G1 = (V1 , E1 ) and G2 = (V2 , E2 ) with

disjoint vertex sets V1 and V2 , denoted by G1 G2 , is a graph with vertex set
V1 × V2 = {(u, v) | u ∈ V1 , v ∈ V2 }, where the vertices x = (x1 , x2 ) and y =
(y1 , y2 ) are adjacent in G1 G2 if and only if [x1 = y1 and {x2 , y2 } ∈ E2 ] or
[x2 = y2 and {x1 , y1 } ∈ E1 ].
700 M. Alipour nad P. Tittmann

Theorem 3 [5]. Let G1 = (V1 , E1 ) and G2 = (V2 , E2 ) be simple undirected

graphs. Then

N (G1 G2 , x) = 1 + |V1 |(N (G2 , x) − 1) + |V2 |(N (G1 , x) − 1)

X   
+ (1 + x)|NG1 (u)| − 1 (1 + x)|NG2 (v)| − 1
(u,v)∈V (G1 G2 )

− |V1 ||V2 |x − 2|E1 ||E2 |x2 .

2.1. Cut vertices

In this section, we consider connected graphs with cut vertices and prove that
there exists an interesting relation between the neighborhood polynomial of a
graph G with the neighborhood polynomials of its split components.
Let G = (V, E) be a simple undirected graph and let v ∈ V . By G − v
we denote the subgraph of G induced by V \ {v}. A vertex v in a connected
graph G is called a cut vertex (or an articulation) of G if G − v is disconnected.
More generally, a vertex v is a cut vertex of a graph G if v is a cut vertex of a
component of G.
Let v be a cut vertex of the graph G = (V, E) and suppose G − v has two
components. Then we can find two subgraphs G1 = (V1 , E1 ) and G2 = (V2 , E2 )
of G such that

V1 ∪ V2 = V, V1 ∩ V2 = {v}, E1 ∪ E2 = E, E1 ∩ E2 = ∅.

We call the graphs G1 and G2 the split components of G.

Theorem 4. Let G = (V, E) be a simple connected graph where v ∈ V is a

cut vertex of G such that G − v has two components. Let G1 = (V1 , E1 ) and
G2 = (V2 , E2 ) be the split components of G. Then the neighborhood polynomial
of G is

N (G, x) = N (G1 , x) + N (G2 , x) − (1 + x)

  
+ (1 + x)|NG1 (v)| − 1 (1 + x)|NG2 (v)| − 1 .

Proof. Let X be a vertex subset of G with X ∈ N (G), which implies that

the vertices of X have a common neighbor in G. Then we can distinguish the
following cases.
(a) Assume X ⊆ V1 or X ⊆ V2 . Then all possibilities for the selection of
X are generated by the polynomial N (G1 , x) + N (G2 , x) − (1 + x), in which we
prevent the double-counting of the empty set and the vertex v by subtracting
1 + x.
Graph Operations and Neighborhood Polynomials 701

(b) Now assume that X contains at least one vertex from each of V1 \ {v} and
V2 \ {v}. In this case, v is the common neighbor, so we need to count all subsets
of the open neighborhood of v in G which include at least one vertex from each of
the open neighborhoods of v in G1 and G2 , which is performed by the generating
function ((1 + x)|NG1 (v1 )| − 1)((1 + x)|NG2 (v2 )| − 1).

Let G = (V, E) be a simple undirected connected graph. A vertex cut (or a

separator ) is a set of vertices of G which, if removed together with any incident
edges, makes that the remaining graph is disconnected. More generally, a vertex
subset W is a vertex cut if W is a vertex cut of a component of G.
Let W ⊆ V be a vertex cut of G which is an independent set in G such that
G − W has two components, where G − W denotes the subgraph of G induced by
V \ W . We can find two subgraphs G1 = (V1 , E1 ) and G2 = (V2 , E2 ) of G such
that
V1 ∪ V2 = V, V1 ∩ V2 = W, E1 ∪ E2 = E, E1 ∩ E2 = ∅.
We call the graphs G1 and G2 the split components of G. In the following theorem
we prove that similar to Theorem 4, there is a relation between the neighborhood
polynomial of a graph containing a vertex cut and the neighborhood polynomials
of its split components.
Theorem 5. Let G = (V, E) be a simple connected graph, W ⊆ V a vertex cut
with |W | = k that is also an independent set in G such that G − W has two
components. Let G1 = (V1 , E1 ) and G2 = (V2 , E2 ) be the split components of G.
Then
X
N (G, x) = N (G1 , x) + N (G2 , x) − AU
U ⊆W
X  T  T 
|U |+1 |
+ (−1) (1 + x) u∈U NG1 (u)|
− 1 (1 + x)| u∈U NG2 (u)| − 1
U ⊆W
U 6=∅

where
x|U | if [U = ∅] or
( T T 
u∈U NG1 (u) 6= ∅ and u∈U NG2 (u) 6= ∅ ,
AU =
0 otherwise.
Proof. Let X be a vertex subset of G with X ∈ N (G), which implies that
the vertices of X have a common neighbor in G. Then we can distinguish the
following cases.
(a) If X ⊆ V1 or X ⊆ V2 , all possibilities to select X are generated by the
polynomial X
N (G1 , x) + N (G2 , x) − AU ,
U ⊆W
702 M. Alipour nad P. Tittmann

in which we avoid any subset of the set W having common neighbors in Pboth split
components G1 and G2 , to be counted more than once by subtracting U ⊆W AU .
(b) Now assume that X contains at least one vertex from each of V1 \ W and
V2 \ W . In this case the common neighbor is in W . We generate all subsets of
the open neighborhoods of vertices of W by the generating function
X  T  T 
(−1)|U |+1 (1 + x)| u∈U NG1 (u)| − 1 (1 + x)| u∈U NG2 (u)| − 1 ,
U ⊆W
U 6=∅

where by applying the principle of inclusion-exclusion we prevent any double

counting.

2.2. Matching edge cuts

Theorem 6. Let F = {{a, b} | a ∈ A, b ∈ B} be a minimal edge cut of a
connected graph G = (V, E) such that F is a matching of G. The components of
G − F are denoted by H and K. Assume that A ⊆ V (H) and B ⊆ V (K). Let
H ′ = H ∪ (A ∪ B, F ) and K ′ = K ∪ (A ∪ B, F ). Then

N (G, x) = N (H ′ , x) + N (K ′ , x) − 2|F |x − 1.

Figure 1 illustrates the situation of Theorem 6.

B
A F

Figure 1. A graph with a matching cut.

Proof. The open neighborhoods of all vertices of V (H) are the same in G and
in H ′ . We also have NG (v) = NK ′ (v) for any v ∈ V (K). This implies

N (G) = N (H ′ ) ∪ N (K ′ ).

Each singleton {v} for v ∈ A ∪ B is contained in N (H ′ ) and in N (K ′ ), which

yields N (H ′ ) ∩ N (K ′ ) = {{v} | v ∈ A ∪ B} ∪ {∅}.
Graph Operations and Neighborhood Polynomials 703

2.3. Edge addition

Let G = (V, E) be a graph, and u, v ∈ V such that {u, v} ∈ / E. Let G + uv =
(V, E ∪ {{u, v}}) be the graph obtained from G by insertion of the edge {u, v}. In
this case, the neighborhoods of the two vertices u, v have been changed. If there
is a path of length 3 between the vertices u and v in G, after the edge insertion,
a 4-cycle is being formed which causes some difficulties in counting neighborhood
sets.
In Theorem 7, we suppose that there is no path of length 3 between the
vertices u and v to prevent any over-counting and relate the neighborhood poly-
nomial of the new graph after addition the edge {u, v} to the neighborhood
polynomial of the original graph. In Theorem 8, we investigate the general case.
Theorem 7. Let G = (V, E) be a graph, u, v ∈ V , and {u, v} ∈ / E. Suppose there
′
is no path of length 3 between u and v in G. Let G = (V, E ∪ {{u, v}}). Then
   
N (G′ , x) = N (G, x) + x (1 + x)|NG (u)| − 1 + x (1 + x)|NG (v)| − 1 .

Proof. Observe that any vertex subset with a common neighbor in G forms a
vertex subset with a common neighbor in G′ . Such vertex subsets are generated
by N (G, x). The addition of the edge {u, v} to G results in the fact that any
non-empty subset of the open neighborhood of u together with v forms a vertex
subset with u as the common neighbor in G′ and the same arguments applies to
the open neighborhood of v. Such subsets are generated by x((1 + x)|NG (u)| −
1) + x((1 + x)|NG (v)| − 1).
Now, assume there is a vertex subset W with a common neighbor in G′ and
suppose it is counted once in x((1 + x)|NG (u)| − 1) + x((1 + x)|NG (v)| − 1) and
once in N (G, x), and let y ∈ W . This implies that W , which is a subset of the
open neighborhood of one of the vertices u (or v) together with v (or u) have
a common neighbor in G. Let x be that common neighbor. The existence of
such set results in the existence of a path of length 3 between u1 and u2 through
vertices x and y which contradicts the assumption that there is no path of length
3 between those two vertices in G. So there is no vertex subset with a common
neighbor in G′ that is counted more than once and this completes the proof.

Theorem 8. Let G = (V, E) be a graph, u1 , u2 ∈ V , and {u1 , u2 } ∈ / E. Suppose

u1 and u2 are not isolated vertices and let G′ = (V, E ∪ {{u1 , u2 }}). Then
X X
N (G′ , x) = N (G, x) + x x|U1 | + x x|U2 | .
∅6=U1 ⊆NG (u1 ) ∅6=U2 ⊆NG (u2 )
NG (U1 )∩NG (u2 )=∅ NG (U2 )∩NG (u1 )=∅

Proof. If X ∈ N (G), then X ∈ N (G′ ) and all such vertex subsets are generated
by N (G, x). After insertion of the edge {u1 , u2 }, the vertices in any subset of the
704 M. Alipour nad P. Tittmann

open neighborhood of u1 together with u2 have a common neighbor in G′ (which

is the vertex u1 ), and analogously the vertices in any subset of the open neigh-
borhood of u2 together with u1 share a neighbor in G′ (which is the vertex u2 ).
Any of such vertex sets adds a term like xx|U | to the neighborhood polynomial
of G′ , where x represents u1 or u2 , and x|U | stands for U as a subset of NG (u2 )
or NG (u1 ), respectively.
In order to prevent counting a vertex set of this form that already has a
common neighbor in G (and therefore being counted in N (G, x)), we consider
only those subsets of NG (u2 ) or NG (u1 ) which do not have a common neighbor
with u1 or u2 in G, respectively. All such vertex subsets are generated by the
term X X
x x|U1 | + x x|U2 | ,
∅6=U1 ⊆NG (u1 ) ∅6=U2 ⊆NG (u2 )
NG (U1 )∩NG (u2 )=∅ NG (U2 )∩NG (u1 )=∅

which completes the proof.

2.4. Vertex attachment

Suppose G = (V, E) is a simple graph and let U ⊆ V . Let

GU ⊲v = (V ∪ {v}, E ∪ {{u, v} | u ∈ U }),

be the graph obtained from G by adding a new vertex v to V and attaching v

to all vertices of U , so the degree of v in GU ⊲v is |U |; in this case the set U
is called the vertex attachment set. In case of U = ∅, the graph GU ⊲v is the
disjoint union of G and the single vertex v, which implies N (GU ⊲v ) = N (G) and
therefore N (GU ⊲v , x) = N (G, x). In the following theorem, we investigate the
neighborhood polynomial of GU ⊲v for any U ⊆ V .

Theorem 9. Let G = (V, E) be a graph and U ⊆ V . For each vertex subset

W ⊆ U , we define a monomial AW by
( |W | T
x if w∈W NG (w) = ∅,
AW =
0 otherwise.

Then
X T X
N (GU ⊲v , x) = N (G, x) + (−1)|W |+1 x(1 + x)| w∈W NG (w)|
+ AW .
W ⊆U W ⊆U
W 6=∅ W 6=∅

Proof. Assume X ∈ N (GU ⊲v ), which implies that the vertices of X have a

common neighbor in GU ⊲v . Then we can distinguish the following cases.
Graph Operations and Neighborhood Polynomials 705

(a) Assume v ∈/ X and v is not a common neighbor of the vertices in X.

In this case X ∈ N (G) and all possibilities for the choice of X are counted in
N (G, x).
(b) Assume v ∈/ X but v is a common neighbor of the vertices in X. Since
the only neighbors of v are the vertices in U , then we only need to count those
subsets of U which do not have a common neighbor in G. We do so by
X
AW ,
W ⊆U
W 6=∅

where (
x|W |
T
if w∈W NG (w) = ∅,
AW =
0 otherwise.

(c) Finally assume v ∈ X. In this case, we need to count all vertex subsets
of the open neighborhoods of vertices u ∈ U where each one of those subsets
together with v forms a vertex subset, with u as their common neighbor in GU ⊲v .
To exclude double counting subsets of intersections of the neighborhoods of u’s,
we use the principle of inclusion-exclusion. Such subsets are generated by
X T
(−1)|W |+1 x(1 + x)| w∈W NG (w)|
,
W ⊆U
W 6=∅

where the factor x accounts for the vertex v itself.

Finally, the arguments in (a), (b), and (c) together prove the theorem.

In the following corollary we rephrase Theorem 9 in a way that the neigh-

borhood polynomial of a graph resulting from vertex removal (instead of vertex
attachment) is investigated.

Corollary 10. Let G = (V, E) be a graph and v ∈ V . Then we have

X T X
N (G, x) = N (G − v, x) + (−1)|U |+1 x(1 + x)| u∈U NG−v (u)|
+ AU ,
U ⊆NG (v) U ⊆NG (v)
U 6=∅ U 6=∅

where for U ⊆ NG (v)

(
x|U |
T
if u∈U NG−v (u) = ∅,
AU =
0 otherwise.
706 M. Alipour nad P. Tittmann

3. Graph Degeneracy

Corollary 10 suggests a recursion for the neighborhood polynomial of a graph

based on removing the vertices of the graph one by one. In this section, we in-
troduce k-degenerate graphs in which the calculation of the neighborhood poly-
nomial is efficiently possible using the mentioned recursion.
A simple undirected graph G = (V, E) of order n is called k-degenerate if
every non-empty subgraph of G, including G itself, has at least one vertex of
degree at most k for 0 < k < n. Suppose G = (V, E) is a k-degenerate graph of
order n. To apply Corollary 10 we need to select a vertex in G to remove.
Since G is k-degenerate, the existence of a vertex of degree at most k is
guaranteed. Suppose v1 is a vertex of degree k in G, then we remove it and by
using Corollary 10, we have
X T
N (G, x) = N (G − v1 , x) + (−1)|U |+1 x(1 + x)| u∈U NG−v1 (u)|
U ⊆NG (v1 )
U 6=∅
X
+ AU .
U ⊆NG (v1 )
U 6=∅

We define Gn := G where n is the number of vertices of G. For each i, 0 < i < n,

let Gn−i be the graph obtained from Gn−(i−1) after removing a vertex vi of
degree at most k. The existence of such vertex is guaranteed by k-degeneracy of
the graph G. We define for every non-empty U ⊆ NGn−(i−1) (vi )
(
x|U | if
T
i u∈U NGn−i (u) = ∅,
AU =
0 otherwise.

and for each i, 0 < i < n

h T i
(−1)|U |+1 x(1 + x)| u∈U NGn−i (u)| + AiU .
X
Xi =
U ⊆NGn−(i−1) (vi ),U 6=∅

Then we have
N (Gn , x) = N (Gn−1 , x) + X1
= (N (Gn−2 , x) + X2 ) + X1
..
.
= (· · · ((N (G1 , x) + Xn−1 ) + Xn−2 ) + · · · ) + X1 )
n−1
X
=1+ Xi ,
i=1
Graph Operations and Neighborhood Polynomials 707

where the last equation is a result of the fact that G1 is equal to a single vertex
which has neighborhood polynomial 1.
Clearly, in each step (in other words for each i, 0 < i < n), we remove a
vertex vi of degree at most k and calculate the term Xi that is a sum over all
non-empty subsets of NGn−(i−1) (vi ), but since this set has at most k elements
due to the degree of vi in Gn−(i−1) , the calculation of N (G, x) can be done in
polynomial time for any fixed k which yields the following theorem.
Theorem 11. Let k be a fixed positive integer. The calculation of the neighbor-
hood polynomial can be performed in polynomial-time in the class of k-degenerate
graphs.
The recursive procedure for the calculation of the neighborhood polynomial
becomes for graphs with regular structure especially simple. We give here an
example of a 2 × n-grid graph that can easily be generalized to similar graphs
with regular structure.
Example 12. A ladder graph of order 2n, denoted by Ln , is the Cartesian prod-
uct of two path graphs P2 and Pn , see Figure 2. The neighborhood polynomial
of Ln is

N (Ln , x) = 1 + 4x + 2x2 + (2n − 4)x(1 + x)2 for n ≥ 2.

v
1 2 3 n

Figure 2. A ladder graph Ln .

Proof. To prove this equation we can calculate the neighborhood polynomial

of Ln by applying Theorem 3, but since Ln is 2-degenerate, we can also apply
Corollary 10. To do so, we need to specify a vertex to remove. In the following,
we remove a vertex of degree 2, denoted by v in Figure 2. After removing the
vertex v, we obtain a modified ladder graph that we denote by Mn , see Figure
3. The graph Mn has exactly one vertex of degree 1. This is the vertex to be
removed in the next step; it is denoted by v in Figure 3. By Corollary 10, we
obtain

(3) N (Ln , x) = N (Mn , x) + x(1 + x)2 for n ≥ 3

and

(4) N (Mn , x) = N (Ln−1 , x) + x(1 + x)2 for n ≥ 3.

708 M. Alipour nad P. Tittmann

1 2 3 n-1

Figure 3. A modified ladder graph Mn .

Substituting N (Mn , x) in Equation (3) according to Equation (4) yields

N (Ln , x) = N (Ln−1 , x) + 2x(1 + x)2 for n ≥ 3,

which provides together with the initial value

N (L2 , x) = N (C4 , x) = 1 + 4x + 2x2

the above given result.

As any k-regular graph is k-degenerate, we obtain the following result.

Corollary 13. Let k be a fixed positive integer. The calculation of the neigh-
borhood polynomial can be performed in polynomial-time in the class of k-regular
graphs.

As a consequence of Euler’s polyhedron formula, each simple planar graph

contains a vertex of degree at most 5. This implies that any simple planar graph
is 5-degenerate, which provides the next statement.

Corollary 14. The neighborhood polynomial of a simple planar graph can be

found in polynomial-time.

There is an interesting generalization of planar graphs which was introduced

in [9] that belongs to the class of k-degenerate graphs, too. An almost planar
graph is a non-planar graph G = (V, E) in which for every edge e ∈ E, at least one
of the graphs G − e (obtained from G after removing e) and G/e (obtained from
G by contraction of e) is planar. It can be easily shown that every finite almost-
planar graph is 6-degenerate, which implies that its neighborhood polynomial can
be efficiently calculated.

4. Complexity

In [11], it is shown that counting dominating sets in several graph classes is #P-
complete, which implies that calculating the domination polynomial and hence,
by Equation 2, the neighborhood polynomial of a graph is #P-hard [14]. For the
problem of counting dominating sets, even parameterized counting is hard. It
Graph Operations and Neighborhood Polynomials 709

has been shown in [8] that counting dominating sets of size k, for a given positive
integer k, is #W[2]-hard.
Consequently, any result that shows that the neighborhood polynomial can be
calculated in polynomial time in some graph class is highly welcome. Brown and
Nowakowski showed in their seminal paper [5] that the neighborhood polynomial
can be calculated in polynomial time in any C4 -free graph. The results given in
[10] show that we can generalize this result to graphs that exclude small complete
bipartite subgraphs. (A cycle C4 can also be considered as a complete bipartite
subgraph K2,2 .) Equation 2 shows that whenever the neighborhood polynomial
can be found in polynomial time for some graph class H, then the domination
polynomial can be calculated efficiently in the class of all complements of graphs
in H.
All positive complexity results found so far concern sparse graphs, as we either
restrict degrees of the graph or exclude some small subgraphs, which corresponds
to the known fact that the domination polynomial can be calculated more effi-
ciently in graphs of high edge density. Hence an interesting open problem is to
find some edge-dense graph classes that allow a polynomial-time computation of
the neighborhood polynomial.

5. Conclusions and Open Problems

The presented decomposition and reduction methods for the calculation of the
neighborhood polynomial work well for graphs of bounded degree or, more gen-
erally, for k-degenerate graphs. They can easily be extended to derive the
neighborhood polynomials of further graphs with regular structure such as grid
graphs with additional diagonal edges. The splitting formula for vertex separa-
tors also suggests that the neighborhood polynomial should be polynomial-time
computable in the class of graphs of bounded treewidth.
A main open problem remains the calculation of the neighborhood polyno-
mial of graphs for which the number of edges is not linearly bounded by its order.
Can we find an efficient way to calculate the neighborhood polynomial of graphs
of bounded clique-width?
For some graph polynomials, like the matching polynomial, we can find nice
relations between the corresponding polynomial of a graph G and the polynomial
of its complement Ḡ. Does there exist a similar relation for the neighborhood
polynomial?
For a given graph G = (V, E), a nice approach to calculate a graph polynomial
is splitting along a separating vertex set, say U ⊆ V . This method has been
successfully applied to the Tutte polynomial, see [2], from which we easily obtain
similar splitting formulae for chromatic, flow, and reliability polynomials as well.
In this paper, we proved a splitting formula for the special case that one of the
710 M. Alipour nad P. Tittmann

split components with respect to the separator U is a star (the vertex addition
formula). However, the main obstacle to generalize this idea is that we cannot
restrict the consideration to subsets of the separator, but also all neighborhoods
of those sets. Is there a general splitting formula for the neighborhood polynomial
that avoids this difficulty?
Acknowledgements
We thank two anonymous reviewers who provided helpful comments on earlier
drafts of the manuscript.
This work was partially supported by the European Social Fund (ESF).

References

[1] S. Akbari, S. Alikhani and Y.-h. Peng, Characterization of graphs using domination
polynomials, European J. Combin. 31 (2010) 1714–1724.
doi:10.1016/j.ejc.2010.03.007
[2] A. Andrzejak, Splitting formulas for Tutte polynomials, J. Combin. Theory Ser. B
70 (1997) 346–366.
doi:10.1006/jctb.1997.1767
[3] J.L. Arocha and B. Llano, Mean value for the matching and dominating polynomials,
Discuss. Math. Graph Theory 20 (2000) 57–69.
doi:10.7151/dmgt.1106
[4] A.E. Brouwer, P. Csorba and A. Schrijver, The number of dominating sets of a finite
graph is odd (2009).
https://www.win.tue.nl/∼aeb/preprints/domin4a.pdf
[5] J.I. Brown and R.J. Nowakowski, The neighbourhood polynomial of a graph, Aus-
tralas. J. Combin. 42 (2008) 55–68.
[6] M. Dod, The independent domination polynomial.
arXiv preprint arXiv:1602.08250 (2016).
[7] K. Dohmen and P. Tittmann, Domination reliability, Electron. J. Combin. 19 (2012)
#P15.
doi:10.37236/1166
[8] J. Flum and M. Grohe, The parameterized complexity of counting problems, SIAM
J. Comput. 33 (2004) 892–922.
doi:10.1137/S0097539703427203
[9] B.S. Gubser, A characterization of almost-planar graphs, Combin. Probab. Comput.
5 (1996) 227–245.
doi:10.1017/S0963548300002005
[10] I. Heinrich and P. Tittmann, Neighborhood and domination polynomials of graphs,
Graphs Combin. 34 (2018) 1203–1216.
doi:10.1007/s00373-018-1968-7
Graph Operations and Neighborhood Polynomials 711

[11] S. Kijima, Y. Okamoto and T. Uno, Dominating set counting in graph classes,
in: Proc. 17th Annual International Conference, COCOON 2011, Lecture Notes in
Comput. Sci. 6842 (2011) 13–24.
doi:10.1007/978-3-642-22685-4 2
[12] T. Kotek, J. Preen, f. Simon, P. Tittmann and M. Trinks, Recurrence relations and
splitting formulas for the domination polynomial, Electron. J. Combin. 19 (2012)
#P47.
doi:10.37236/2475
[13] T. Kotek, J. Preen and P. Tittmann, Subset-sum representations of domination
polynomials, Graphs Combin. 30 (2014) 647–660.
doi:10.1007/s00373-013-1286-z
[14] T. Kotek, J. Preen and P. Tittmann, Domination polynomials of graph products.
arXiv preprint arXiv:1305.1475 (2013)
[15] L. Lovász, Kneser’s conjecture, chromatic number, and homotopy, J. Combin. The-
ory Ser. A 25 (1978) 319–324.
doi:10.1016/0097-3165(78)90022-5

Received 5 December 2019

Revised 7 July 2020
Accepted 7 July 2020